A language description, such as a grammar specification, is required to 
construct a tool that performs analysis, comprehension, maintenance, 
or visualization tasks for programs written in the language. However, 
a language dialect, which is a variation and/or extension of a language, 
is typically not accompanied by a grammar specification and is sometimes 
not accompanied by either a language manual or an ad hoc specification.  
Therefore, the only representation of the grammar for the language dialect 
frequently is the programmatic representation of the grammar found in the 
source code of a compiler for the language.  This is problematic because 
parser generation algorithms place restrictions on the form of a compiler 
grammar, thereby making it difficult to comprehend.  Nevertheless, there 
has been little work on recovering a grammar specification for a language 
or a language dialect from existing language artifacts.  Furthermore, there 
exists no methodology that enables a language engineer to automatically 
recover a grammar specification for a language.

In this paper we present a methodology for automated grammar recovery from 
existing language artifacts; in particular, the methodology takes as input 
a set of parse tree instances.  We then describe a system that implements 
the methodology and present empirical studies that we performed using the 
system.  Our studies encompass the extraction, measurement, assessment, 
refactoring, and exchange of a grammar for a language dialect. We focus on 
the GNU dialect of C++ in our studies, but our methodology is applicable to 
other well-known language dialects, including Eclipse C/C++ and Jikes Java.